Digital Twins in Action: AI, HPC, and End-to-End Workflows at Scale

Digital twins—computational representations of physical systems that evolve with data—are emerging as a powerful paradigm in healthcare for modeling disease progression, predicting treatment response, and enabling personalized interventions. Realizing this vision requires more than accurate models; it depends on scalable, end-to-end workflows that integrate data, simulation, and artificial intelligence across the AI–HPC continuum.

This talk presents a systems-level view of digital twin workflows, emphasizing the architectural and computational requirements for operating at scale. It outlines an end-to-end framework that connects multimodal data ingestion, preprocessing, and harmonization with physics-based simulation and machine learning, executed on high-performance computing platforms. Particular focus is placed on workflow orchestration, modular design, and user-facing interfaces that enable reproducibility and accessibility without exposing underlying infrastructure complexity.

Representative applications in medical imaging and radiation science are used to illustrate these concepts, including population-scale virtual imaging trials and patient-specific dosimetry simulations. These examples demonstrate how integrated workflows enable controlled, reproducible studies and generate structured datasets for model development and evaluation, while also highlighting challenges in data integration, portability, and performance.

The talk concludes with key lessons learned in operationalizing digital twins across the AI–HPC continuum. These include the importance of workflow-centric design, standardized data representations, and the complementary roles of HPC and AI in enabling predictive, data-driven science. The approaches discussed provide a foundation for scalable digital twin systems applicable across scientific domains.